A SkillsCast for this session is not available.
Support vector machine remains a widely used classical machine learning algorithm in practice. We have ported liblinear's L2-regularised hinge loss binary classification routine to Rust and packaged it as a pip-installable Python package. Lisbon is a drop-in replacement for sci-kit-learn's SVC routine three and reproduces the original library's results. By changing matrix representation from sparse to dense, Lisbon has achieved a 20-40% speedup in computational time and a greater than 50% memory saving by eliminating memory copies. In this talk, I will present the key findings from this work and the lessons learnt during the process.
YOU MAY ALSO LIKE: